How I Built an AI That Breeds Its Own Jailbreaks Using Genetic Algorithms

A researcher created Basilisk, an open-source AI security tool that uses genetic algorithms to automatically evolve adversarial prompts that can bypass AI safety filters, replacing static jailbreak lists that become obsolete when models get patched. The system treats prompts like organisms that mutate and reproduce, achieving a 92% improvement in attack success rates by generation 5.